Florida Implantable Wireless Recording Electrodes 
University of Florida: F.W.I.R.E.

Along with several lab mates and under the supervision of the PI Dr. John G. Harris, we are working on the design, fabrication and implementation of a fully implantable wireless chip for neural data recording in current Brain Machine Interfaces (BMI). The signals of interest are neuron action potentials recorded using micro-electrodes. Due to the number of recording channels and the bandwidth of the signal, the data rates are unmanageable with conventional sampling schemes (specially using a wireless link). The power, area and bandwidth constraints can be met using custom encoding schemes (event-based, input dependent samplers). One example, is the well known integrate-and-fire model, which has been used to encode neural recordings at sub-Nyquist data rates while providing an accurate recovery (signal to error ratio 50dB) in the action potential region. These sampling schemes provide simple hardware front-ends and  reduce the data rates by exploiting the localized features of the neural recordings. However, linear filters will no longer recover the input and therefore  elaborate recovery algorithms are needed. Nevertheless, there are sufficient resources at the receiver side to deal with these requirements. 

HP Open Innovation Office program
University of Florida / HP Laboratories 
Collaborators: Evan Kriminger (UF) and Dr. Choudur Lakshminarayan (HPL)


Anomaly Detection in Multivariate Data Streams using Kernel Methods and Information Theoretic Cost Functions: Focusing on real time algorithms for streaming data designed to predict and model anomalies and pattern in large scale data sets.

Integrate-and-Fire Sampling and Reconstruction Using a Memristor-Based Sampler for Massive Sensor Networks 
University of Florida / HP Laboratories :
Collaborators: Manu Rastogi (UF) and Dr. Choudur Lakshminarayan (HPL)

We are still at a preliminary state with this project. The basic idea is to implement efficient adaptive sampling schemes into massive sensor networks. Given that size is a major factor, the samplers have been developed using memristors.